Journal of Homotopy and Related Structures, vol. 1(1), 2006, pp.195–218
ALGEBRAIC K-THEORY OF FREDHOLM MODULES AND
KK-THEORY
TAMAZ KANDELAKI
(communicated by Jonathan Rosenberg)
Abstract
Kasparov KK-groups KK(A, B) are represented as homotopy groups of the Pedersen-Weibel nonconnective algebraic Ktheory spectrum of the additive category of Fredholm (A, B)bimodules for A and B, respectively, a separable and σ-unital
trivially graded real or complex C ∗ -algebra acted upon by a
fixed compact metrizable group.
dedicated to Hvedri Inassaridze
Introduction
In noncommutative topology and differential geometry some of the useful and
powerful tools are methods of algebraic K-theory, Kasparov’s KK-theory, spectra
and so on. Therefore a comprehensive study of relationships between them may be
considered as an interesting task. The main goal of this paper is characterization of
Kasparov KK-groups as algebraic K-groups of an additive category. On first view,
calculation of the KK-theory by algebraic K-theory seems to be highly improbable,
as algebraic K-theory and KK-theory are independent, highly nontrivial, theories,
having almost no connections with each other. The key is thus to find suitable
objects which make sense for both algebraic K-theory and KK-theory. In this
paper we concentrate on the additive C ∗ -category Rep(A, B), namely the category
of Fredholm modules, where A is a separable and B is a σ-unital real or complex
C ∗ -algebra with action of a fixed compact second countable group. Our main result
claims the natural isomorphisms
Kan (Rep(A; B)) ' KKn−1 (A; B),
where Kan denote the algebraic K-functors isomorphic to Quillen’s K-functors in
nonnegative dimensions, and isomorphic to Pedersen-Weibel K-functors in negative
dimensions.
There are already several papers dedicated to interpretations of KK-theory, each
with their own advantages. Let us point out in brief some fundamental papers of this
Received May 5, 2006, revised August 30, 2006; published on September 19, 2006.
2000 Mathematics Subject Classification: 19K35, 19K33, 19D23, 19D99, 46L99.
Key words and phrases: Nonconnective algebraic K-theory spectrum, category of Fredholm modules, KK-theory.
c 2006, Tamaz Kandelaki. Permission to copy for private use granted.
°
Journal of Homotopy and Related Structures, vol. 1(1), 2006
196
sort which are sources of further research. These are G. Kasparov’s interpretation
of KK-theory in terms of extensions of C ∗ -algebras [24]; J. Cuntz’s result in [6],
based on the homotopy category of C ∗ -algebras; N. Higson’s approach, considering
KK-theory as an universal enveloping additive category of the category of separable
C ∗ -algebras [6]; the interpretation of the KK-theory with the aid of the category
of C ∗ -algebras and asymptotic homomorphisms, due to A. Connes and N. Higson
[4], and R. Meyer’s and R. Nest’s joint paper [25], where Kasparov category KK
turned into a triangulated category.
Let us shortly review some known results in KK-theory established by methods
of topological K-theory and spectra.
In [29] E. K. Pedersen and C. Weibel showed that values on finite CW -complexes
X of the homology theory associated with the nonconnective algebraic K-theory
spectrum of a unital ring R may be interpreted as the algebraic K-groups (up to a
shift in dimension) of a suitably constructed additive category CO(X) (R). According
to this result, in [31] J. Rosenberg showed that
K1 (CO(X) (R)) ' KK(C(X), R)
for a unital C ∗ -algebra R. He also constructed algebraic KK-theory spectra, denoted by KK(A, B), having the property
π0 (KK(A, B)) ' KK(A, B).
The similar question for nonzero dimensions has been left open in that paper.
In [12], there have been constructed KKtop (C0 (X), B)-spectra, related to Kasparov groups KKn (C0 (X), B), which were used to construct the splitting assembly
map
A : KKtop (C0 (X), B) → Ktop (C ∗ (X)),
where C ∗ (X) is the C ∗ -algebra of the coarse space X. A large number of results on
the Novikov conjecture can be included under this scheme [12].
More general approaches to non-equivariant KK-theory as homotopy groups of a
spectrum can be found in the following papers. These are, [13], where
T. G. Houghton-Larsen and K. Thomsen, utilizing spaces of C ∗ -extensions, have
constructed KK-theory spectra; and [26], where P. Mitchener, using methods of
topological K-theory, symmetric spectra and C ∗ -categories has defined KK-theory
spectra, too.
Let us make few remarks concerning our approach. In [16], [19] we have calculated topological and Karoubi-Villamayor K-groups of the C ∗ -category Rep(A, B)
which are related to KK-groups by the equalities
Knt (Rep(A, B)) = KnKV (Rep(A, B)) = KKn−1 (A, B), n > 0,
∗
(0.1)
where A and B are G-C -algebras, separable and σ-unital respectively. Thus the
additive C ∗ -category Rep(A, B)) is a good object for our purposes at first sight.
In the article [17] it was announced that an isomorphism similar to 0.1 for algebraic K-groups also holds, as well. The present paper is an attempt to explain in
detail the results announced in [17]. As a consequence construction of a nonconnective algebraic KK-theory spectrum KKalg (A, B) = K(Rep(A, B)) arises, where the
Journal of Homotopy and Related Structures, vol. 1(1), 2006
197
right hand side is the Pedersen-Weibel nonconnective algebraic K-theory spectrum
of the idempotent-complete additive category Rep(A, B). This spectrum has the
property
πn (KKalg (A, B)) ' KKn−1 (A, B),
for all n ∈ Z. In a further paper we hope to apply the algebraic KK-theory spectra,
in particular, to problems related to the Novikov conjecture. Our approach is mainly
based on the author’s unpublished preprint [20].
1.
The main Theorem and an outline of the proof
The purpose of this section is to summarize some of the concepts that are needed
for the formulation of the main theorem, and then to give an outline of the proof.
First, we recall the definitions of a C ∗ -category and an idempotent-compete C ∗ category; and in Subsection 1.1 an idempotent-complete C ∗ -category Rep(A, B) is
constructed that we will need later on.
Let A be a category such that for any pair (a, b) of objects in A, the set hom(a, b)
is equipped with the structure of a Banach space in such a way that composition
is a continuous k-bilinear map. Such a category is said to be a Banach category
over k, or simply a Banach category. A Banach category A is called a C ∗ -category
if it is equipped with a family of anti-linear maps ∗ : hom(a, b) → hom(b, a) for any
a, b ∈ Ob(A) such that
1. (f ∗ )∗ = f ;
2. (f g)∗ = g ∗ f ∗ , if f g exists;
3. kf ∗ k = kf k;
4. kf ∗ f k = kf k2 , if k is the complex numbers; and kf k2 6 kf ∗ f + g ∗ gk2 , if k is
the real numbers.
5. For any morphism f : a → b in A the morphism f ∗ f is a positive element of
the C ∗ -algebra hom(a, a).
Let A and B be C ∗ -categories. A functor F : A → B is said to be a ∗-functor if
• F(f + g) = F(f ) + F(g);
• F(λf ) = λF(f );
• F(f ∗ ) = F ∗ ,
where λ ∈ k, and f and g are morphisms in A. (cf. [3], [16], [18]).
We say that a ∗-functor is faithful if it is injective on both objects and morphisms.
Any ∗-functor is norm-nonincreasing. Moreover, a faithful ∗-functor preserves norms
[3].
The category H(k) of separable Hilbert spaces and bounded linear maps has
a natural structure of a C ∗ -category. There exists a faithful ∗-functor from every
C ∗ -category into H(k).
Let A be a C ∗ -category and I ⊂ hom A. Put homI (a, b) = hom(a, b)∩I. Then I is
called a left ideal if homI (a, b) is a linear subspace of hom(a, b) and f ∈ homI (a, b),
g ∈ hom(b, c) imply gf ∈ homI (a, c). A right ideal is defined similarly. I is a twosided ideal if it is both a left and a right ideal. An ideal I is closed if hom(a, b)I
Journal of Homotopy and Related Structures, vol. 1(1), 2006
198
is closed in hom(a, b) for each pair of objects. A closed two-sided ideal I is called
a C ∗ -ideal if I = I ∗ . Every C ∗ -ideal determines an equivalence relation on the
morphisms of A: f ∼ g if f − g ∈ I; the set of equivalence classes A/I can be made
into a C ∗ -category in a unique way by requiring that the canonical map f 7→ fˆ gives
rise to a ∗-functor A → A/I. Arguing as for C ∗ -algebras, one can show that every
closed ideal is a C ∗ -ideal [3]. Among the C ∗ -categories containing I as a C ∗ -ideal
there exists a universal one, the so called multiplier C ∗ -category. We will denote by
M (I) the multiplier C ∗ -category of a C ∗ -ideal I [3].
Let A be an additive category. Idempotent completion of A is an additive category
 whose objects have the form (a, q), where a is an object in A and q is an idempotent in hom(a, a), and a morphism f : (a, q) → (a0 , q 0 ) is a morphism f : a → a0
such that f q = q 0 f = f . There is a natural functor A → Â, defined by assignments
a 7→ (a, 1a ) and f 7→ f . An additive category B is said to be idempotent-complete
if there are an additive category A, and an additive functor F : B → Â which is an
equivalence of categories.
Note that for an additive C ∗ -category A the category  is not necessarily a
∗
C -category. Below we will adapt the above construction to the case of additive
C ∗ -categories.
Recall that a projection p in a C ∗ -category is a morphism with the properties
∗
p = p and p2 = p, i. e., a projection is a selfadjoint idempotent.
Let A be an additive C ∗ -category. Consider the additive C ∗ -category à with
objects of the form (a, p), where a ∈ Ob(A) and p ∈ hom(a, a) is a projection. A
morphism from (a, p) to (b, q) is a morphism f : a → b in A such that f p = qf = f .
Composition of morphisms is the same as in A. The sum is given by (a, p) ⊕ (b, q) =
(a⊕b, p⊕q), and the norm of morphisms is inherited from A [16]. There is a natural
functor ν : Ã → Â defined by identity maps on objects and morphisms.
Let us show the following simple lemma.
Lemma 1.1. Let A be an additive C ∗ -category. Then à is an idempotent-complete
C ∗ -category.
Proof. Consider the natural additive functor ν : Ã → Â, which is, of course, faithful.
Let us show that ν is a full functor. Indeed, if q ∈ hom(a, a) is an idempotent then
1
1
p = ((2q ∗ − 1)(2q − 1) + 1) 2 · q · ((2q ∗ − 1)(2q − 1) + 1)− 2
is a projection and the pairs (a; q) and (a; p) are isomorphic in  via the morphism
1
p((2q ∗ − 1)(2q − 1) + 1) 2 q.
Now, we define examples of additive C ∗ -categories which are used in the remaining part of paper.
1.1. On additive C ∗ -categories Rep(A, B) and Rep(A, B)
Let HG (B) be the additive C ∗ -category of countably generated right Hilbert Bmodules equipped with a B-linear, norm-continuous G-action over a fixed compact
second countable group G [23]. Note that the compact group acts on the morphisms
Journal of Homotopy and Related Structures, vol. 1(1), 2006
199
by the following rule: for f : E → E 0 the morphism gf : E → E 0 is defined by the
formula (gf )(x) = g(f (g −1 (x))).
The category HG (B) contains the class of compact B-homomorphisms [23]. Denote it by KG (B). Known properties of compact B-homomorphisms imply that
KG (B) is a C ∗ -ideal [3] in HG (B).
Objects of the category Rep(A, B) are pairs of the form (E, ϕ), where E is an
object in HG (B) and ϕ : A → L(E) is an equivariant ∗-homomorphism. A morphism
f : (E, φ) → (E 0 , φ0 ) is a G-invariant morphism f : E → E 0 in HG (B) such that
f φ(a) − φ0 (a)f ∈ KG (E, E 0 )
for all a ∈ A. The structure of a C ∗ -category is inherited from HG (B). It is easy to
see that Rep(A, B) is an additive C ∗ -category, not idempotent-complete.
Now, we are ready to construct our main C ∗ -category, that is Rep(A, B). Objects
of it are triples (E, φ, p), where (E, φ) is an object and p : (E, φ) → (E, φ) is a
morphism in Rep(A, B) such that p∗ = p and p2 = p. A morphism f : (E, φ, p) →
(E 0 , φ0 , p0 ) is a morphism f : (E, φ) → (E 0 , φ0 ) in Rep(A, B) such that f p = p0 f = f .
In detail, f must satisfy
f φ(a) − φ0 (a)f ∈ K(E, F ) and f p = p0 f = f.
(1.1)
So, by definition
^B).
Rep(A, B) = Rep(A,
The structure of a C ∗ -category on Rep(A, B) comes from the corresponding structure on Rep(A, B).
Let SG denote the category of trivially graded separable C ∗ -algebras over k with
an action of the compact second countable group G and equivariant ∗-homomorphisms. Functors Kan are defined by
Kan (A) = πn K(A),
n ∈ Z,
where K(A) is the Pedersen-Weibel nonconnective algebraic K-theory spectrum
[28] of an idempotent-complete additive category A. Functors Ktn are the topological K-functors on idempotent-complete additive C ∗ -categories, defined by Karoubi
−n
[22],[21]. For simplicity, Kasparov’s groups KKG
(A, B) will be denoted by
KKn (A, B).
Now, we present our main result in the following theorem.
Theorem 1.2. Let B be a σ-unital trivially graded C ∗ -algebra with an action of a
second countable compact group G. There are natural isomorphisms
Kan (Rep(−; B)) ' Ktn (Rep(−; B)) ' KKn−1 (−; B)
(1.2)
of functors on the category SG , for all n ∈ Z.
Outline of proof. Theorem 1.2 is a consequence of the argument presented below.
A family H = {Hn }n∈Z of contravariant functors from SG to the category of
abelian groups and homomorphisms is said to be a stable cohomology theory on the
category SG if
Journal of Homotopy and Related Structures, vol. 1(1), 2006
200
1. H has the weak excision property. Namely, for any exact proper sequence
0→I→B→A→0
(which means that the involved epimorphism admits an equivariant completely
positive contractive section) of algebras in SG there exists a natural homomorphism δn : Hn (I) → Hn−1 (A), for any n ∈ Z, such that the resulting natural
sequence of abelian groups (extending in both directions)
δ
n
· · · → Hn (A) → Hn (B) → Hn (I) →
Hn−1 (A) → · · ·
is exact.
2. H is stable. This means that if eA : A → A ⊗ K is a homomorphism defined
by the map a 7→ a ⊗ p, where p is a rank one projection in K, then Hn (eA ) :
Hn (A⊗K) → Hn (A) is an isomorphism, where K is the C ∗ -algebra of compact
operators on a separable Hilbert space H over k, with the trivial action of the
group G.
Denote by Ck (S 1 ) the C ∗ -algebra of continuous complex functions on the standard unit circle S 1 of modulus one complex numbers, in case k is the field of complex
numbers; while if k is the field of reals, let it be the subalgebra of the former consisting of functions invariant under the conjugation defined by the map f (z) 7→ f (z̄).
It is clear that any continuous complex function f : S 1 → C may be represented in
the form
f (z) + f (z̄)
(−if (z)) + (−if (z̄))
f (z) =
+i
,
2
2
which means that CC (S 1 ) is the complexification of the real C ∗ -algebra CR (S 1 ).
Let Tk be the Toeplitz C ∗ -algebra—the universal C ∗ -algebra over k generated
by an isometry v. There is a conjugation on TC defined by the equality v̄ = v on
the generator v of TC . According to the universal property of the Toeplitz algebra,
one gets a natural homomorphism TR → TC so that the induced homomorphism
TR ⊗R C → TC is an isomorphism, so that there is a short exact sequence
t
0 → Kk → Tk → Ck (S 1 ) → 0.
where Kk is the C ∗ -algebra of compact operators on a separable Hilbert space over
k.
Let τ : Ck (S 1 ) → k be the homomorphism given by f 7→ f (1). Denote by fk
the kernel of τ . It is clear that fR is naturally isomorphic to the algebra C0R (iR)
defined in [5], and fC is isomorphic to ΩC , where Ωk = {f : I = [0, 1] → k | f (0) =
f (1) = 0}. The actions of G on the algebras considered above are trivial.
It follows from the Higson’s theorem (see [10], and [15] for the real case) that the
functor Hn is homotopy invariant for any n ∈ Z. Now, the proofs of the following
proposition and theorem coincide (up to trivial changes) with the proofs of suitable
results in [5].
Proposition 1.3. Let H be a stable cohomology theory and let g : Tk → k be the
homomorphism defined by v 7→ 1. Then the homomorphism
'
Hn (idA ⊗ g) : Hn (A) −
→ Hn (A ⊗ Tk )
Journal of Homotopy and Related Structures, vol. 1(1), 2006
201
is an isomorphism, for any A ∈ SG and n ∈ Z.
Theorem 1.4. Let H be a stable cohomology theory, fA = A⊗fk and ΩA = A⊗Ωk .
Then there are natural isomorphisms
Hn+1 (A) ' Hn (fA) and Hn−1 (A) ' Hn (ΩA).
(1.3)
for any A ∈ SG and n ∈ Z.
As a consequence we have the following working principle: two stable cohomology
theories are isomorphic if and only if they are isomorphic in some fixed dimension.
Thus in the remaining sections of the paper we show that the families of functors in
Theorem 1.2 are stable cohomology theories and they are isomorphic when n = 0.
In more detail, in section 2 we study an interpretation of algebraic and topological
K-theories of C ∗ -categories. Our definition is an adaptation to our cases of some
arguments from [1], [9], [30]. Let A be a C ∗ -category and let I be a C ∗ -ideal in A; let
a and a0 be objects in A. We write a 6 a0 if there exists a morphism s : a → a0 such
that s∗ s = 1a (such a morphism is said to be an isometry). Denote by L(a) (resp. by
I(a)) the C ∗ -algebra homA (a, a) (resp homI (a, a)). We have a well-defined inductive
system of abelian groups {Kan (L(a)), σaa0 }a and {Ktn (I(a)), σaa0 }a . We suppose that
a
a
a
Kan (A) = lim
−→a Kn (I(a))).
−→a Kn (L(a)) and Kn (I) = lim
(1.4)
Thanks to the results of A. Suslin and M. Wodzicki on the excision property of
algebraic K-groups on C ∗ -algebras [33], the right hand side of the second equation
is well-defined. Algebraic K-groups obtained in this way are naturally isomorphic to
Quillen’s K-groups KQ
n (A) when n > 0; and are isomorphic to the Pedersen-Weibel
K-groups in negative dimensions. Note that a new interpretation of algebraic Kgroups implies existence of a simple flexible technical tool. Namely, any element of
an algebraic K-group of an additive category may be represented as an element
of an algebraic K-group of the endomorphism algebra of an object, and such an
interpretation is unique up to a manageable equivalence. Throughout the paper,
this principle will be used repeatedly.
In section 2, according to the excision property of algebraic K-groups on the
category of C ∗ -algebras [33], we establish the excision property for a short exact
sequence associated to a C ∗ -ideal in an additive C ∗ -category (see Proposition 2.4).
In section 3 this property is used to prove Theorem 3.1 about the excision property
of functors
{Kan (Rep(−; B))}n∈Z .
(1.5)
In addition to the excision property, proof of Theorem 3.1 uses two nontrivial results.
These are theorem 3.5 and Theorem 3.8.
In section 4, the stability property of the functors 1.5 will be shown.
Now, since the family of Kasparov’s functors KKn (−; B), n ∈ Z is a stable cohomology theory [7], the proof of Theorem 1.2 boils down to showing the isomorphism
K0 (Rep(A, B)) ' KK1 (A, B),
which is done in section 5.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
202
Remark 1.5. Discussions for topological K-groups are omitted, because they literally
coincide with the considered case of algebraic K-groups.
Remark 1.6. The main result (with essential changes in definitions and theorems) is
also true for a locally compact group G. We hope to discuss this case independently,
not in this paper.
2.
Some remarks on the algebraic K-theory of additive C ∗ categories
We will use the Pedersen-Weibel interpretation of algebraic K-groups [28], denoted here by Kan , instead of Quillen’s definition of algebraic K-groups in [30] which
was given through homotopy groups of certain space. In this section we review
some properties of algebraic K-groups of idempotent-complete additive categories,
based on Pedersen-Weibel’s nonconnective spectra (in this context there are defined
negative K-groups, too). Then, we reinterpret algebraic K-groups of idempotentcomplete additive C ∗ -categories and, with the aid of results from [33], generalize
them to C ∗ -ideals in additive C ∗ -categories. This material plays auxiliary role in
this paper.
In the following lemma we list some simple properties of algebraic K-groups
which suffice for our purposes.
Lemma 2.1. Let A be an idempotent-complete small additive category. Then
1. if A = A1 × A2 , then Kan (A) = Kan (A1 ) × Kan (A2 ), n ∈ Z;
2. if {Aα } is a direct system of full additive subcategories in A such that lim
−→Aα =
a
A. Then Kan (A) = lim
K
(A
),
n
∈
Z.
−→ n α
In the remaining part of this section we give an interpretation of the groups
Kan (A), n ∈ Z for additive C ∗ -categories. This interpretation will convenient for our
purposes.
Let a be an object in an idempotent-complete C ∗ -category A. Consider the full
sub-C ∗ -category Aa in A consisting of all those objects a0 of A which admit an
isometry s : a0 → a⊕n . It is clear that Aa is an idempotent-complete C ∗ -category
equivalent to the category P(L(a)) of f. g. projective modules over L(a).
Now, consider a direct system of abelian groups for A based on subcategories
Aa . It is evident that if there exists an isometry s : a0 → a then one has a natural
additive inclusion ∗-functor (not depending on s) ia0 a : Aa0 → Aa and thus we have
the direct system {Aa , ia0 a }(obA,6) of idempotent-complete C ∗ -categories. Because
of the continuity property of algebraic K-groups (property (2) in Lemma 2.1) and
the isomorphism of categories A = lim
−→Aa , one has an isomorphism
a
Kan (A) = lim
−→ Kn (Aa )
n ∈ Z.
This suggests that Kan (A) can be interpreted in the form
a
Kan (A) = lim
−→ Kn (L(a)).
Below it is done in detail.
(2.1)
Journal of Homotopy and Related Structures, vol. 1(1), 2006
203
2.1. Algebraic K-functors of C ∗ -ideals
Let us make some comments on the results of A. Suslin and M. Wodzicki in
algebraic K-theory before we introduce our view on algebraic K-theory of C ∗ -ideals.
One of their main results is, by Proposition 10.2 in [33], that C ∗ -algebras have the
factorization property (TF)right . Thus any C ∗ -algebra possesses the property AHZ .
These results have many useful consequences in algebraic K-theory of C ∗ -algebras,
which are listed below. (Recall that if A is a C ∗ -algebra and A+ is the C ∗ -algebra
obtained by adjoining a unit to A, then
Kna (A) = ker(Kna (A+ ) → Kna (k)),
(2.2)
n ∈ Z).
1. Kia is a covariant functor from the category of C ∗ -algebras and ∗-homomorphisms to the category of abelian groups for any i ∈ Z;
2. For every unital C ∗ -algebra R containing the C ∗ algebra A as a two-sided
ideal, the canonical map Kna (A) → Kna (R, A) is an isomorphism;
3. The natural embedding into the upper left corner A ,→ Mk (A) induces, for
every natural n, an isomorphism Kna (A) ' Kna (Mk (A));
4. Any extension of C ∗ -algebras
0→I→B→A→0
induces a functorial two-sided long exact sequence of algebraic K-groups
a
· · · → Ki+1
(A) → Kia (I) → Kia (B) → Kia (A) → · · ·
(i ∈ Z).
(2.3)
5. Let A be a C ∗ -algebra and let u be a unitary element in a unital C ∗ -algebra
containing A as a closed two-sided ideal. Then the inner automorphism ad(u) :
A → A induces the identity map of algebraic K-groups.
Below we define algebraic K-groups for C ∗ -ideals. These groups possess all properties similar to those represented above.
Let A be an additive C ∗ -category and let J be its closed C ∗ -ideal. Let LA (a) =
homA (a, a) and LA (a, J) = L(a)A ∩ J for any object a ∈ A. The latter is a closed
ideal in the C ∗ -algebra LA (a). Let us write a 6 a0 if there is an isometry v : a → a0 ,
i.e v ∗ v = 1a . The relation “a 6 a” makes the set of objects of A into a directed
system. Any isometry v : a → a0 defines a ∗-homomorphism of C ∗ -algebras
Ad(v) : LA (a) → LA (a0 )
by the rule x 7→ vxv ∗ . It maps LA (a, J) to LA (a0 , J).
Let v1 : a → a0 and v2 : a → a0 be two isometries. Then Adv1 and Adv2 induce
the same homomorphisms
Ad∗ v1 = Ad∗ v2 : Kna (LA (a)) → Kna (LA (a0 ))
and
Ad∗ v1 = Ad∗ v2 : Kna (LA (a, J)) → Kna (LA (a0 , J)).
The similar result for topological K-theory is in [9]. This means that the homo0
0
0
morphism ν∗aa = Kna (ν aa ) is independent of choosing an isometry ν aa : a → a0 .
Journal of Homotopy and Related Structures, vol. 1(1), 2006
204
0
Therefore one has a directed system {Kna (LA (a, J)), ν∗aa )}a,a0 ∈obA of abelian groups,
for all n ∈ Z.
Definition 2.2. Let A be an additive C ∗ -category and let J be its closed C ∗ -ideal.
Define
a
a
a
Kan (A, J) = lim
−→Kn (LA (a, J)) and Kn (J) = Kn (M (J), J)
(2.4)
where M (J) is the so called multiplier C ∗ -category of J [18].
Lemma 2.3. Let J be a C ∗ -ideal in an additive C ∗ -category A. Then
1. Kan (J) = Kan (A, J);
2. if A0 is a cofinal subcategory in A then Kan (A0 ) ' Kan (A);
3. if A is idempotent-complete, then Kan (A) ' Kan (A).
Proof. 1. The natural ∗-functor ρ : A → M (J) induced from the identity on J by
the universal property of M (J) obviously preserves the relation “6”. This implies
that there is a natural morphism of directed systems
0
{ρa }
0
n
→ {Kna (LM (J) (a, J)), ν∗aa )}a,a0 ∈obM (J) ,
{Kna (LA (a, J)), ν∗aa )}a,a0 ∈obA −−−
where the homomorphism ρan : Kna (LA (a, J)) → Kna (LM (J) (a, J)) is induced by the
above ∗-functor ρ : A → M (J). In view of the isomorphism Kn (A) → Kn (R, A) [33]
one concludes that the homomorphism ρan is an isomorphism for all n ∈ Z a ∈ A.
This morphism is cofinal, since if “a 6 a0 ” in M (J) then “a 6 a ⊕ a0 ” in A and
“a0 6 a ⊕ a0 ” in M (J). Therefore, {ρan } is an isomorphism of direct systems.
2. This is an easy consequence of Definition 2.2.
3. This results from comparison of Definition 2.2 and isomorphism 2.1.
Now, we prove the excision property of algebraic K-theory which will be used in
the next section.
Proposition 2.4. Let A be an additive C ∗ -category and let J be a C ∗ -ideal in A.
Then the two-sided sequence of algebraic K-groups
... → Kan+1 (A/J) → Kan (J) → Kan (A) → Kan (A/J) → ...,
(2.5)
n ∈ Z, is exact.
Proof. Consider the exact sequence of C ∗ -algebras
0 → L(a, J) → L(a, A) → L(a, A)/L(a, J) → 0.
By the excision property of algebraic K-theory on C ∗ -algebras [33], one has a twosided long exact sequence of algebraic K-groups
· · · → Kna (L(a, A)/L(a, J)) →
a
a
a
Kn−1
(L(a, J)) → Kn−1
(L(a, A)) → Kn−1
(L(a, A)/L(a, J)) → · · ·
(2.6)
Journal of Homotopy and Related Structures, vol. 1(1), 2006
205
According to the exact sequence 2.6 and the fact that directed colimits preserve
exactness, one obtains the following long exact sequence of abelian groups:
a
· · · → lim
−→Kn+1 (L(a, B)/L(a, J)) →
a
Kan (J) → Kan (B) → lim
−→Kn (L(a, B)/L(a, J)) → · · ·
(2.7)
There is a natural morphism of directed systems
{ωa } : {Kna (L(a, A)/L(a, J)} → {Kna (L(a, A/J)}
so that ωa is the identity map for any object a in A. It is clear that this morphism
is cofinal. Thus the induced homomorphism
a
a
a
ω : lim
−→Kn (L(a, A/J)) = Kn (A/J)
−→Kn (L(a, A)/L(a, J)) → lim
is an isomorphism.
3.
Excision Property of Kan ((Rep(−; B))
In this section we will prove that the contravariant functors Kan ((Rep(−; B)),
n ∈ Z, have the weak excision property. The similar result for topological K-theory,
in a particular case, has been proved in [9].
Remark 3.1. From now on for convenience of calculations the functors
Kan ((Rep(−; B)) are considered instead of Kan ((Rep(−; B)) and Kan ((Rep(−; B)).
Since Rep(A; B) is a cofinal full subcategory in Rep(A; B), by Lemma 2.3 (2),(3)
these functors are isomorphic.
For any closed invariant ideal J in a separable C ∗ -algebra A from SG there is
a C ∗ -ideal D(A, J; B) in Rep(A; B) which is defined in the following manner. Let
(E, φ) and (E 0 , φ0 ) be objects in Rep(A, B). A morphism α : (E, φ) → (E 0 , φ0 ) in
Rep(A, B) is in D(A, J; B) if
αφ(x) ∈ K((E, φ), (E 0 φ0 )) and φ0 (x)α ∈ K((E, φ)), for all x ∈ A.
The space of all morphisms from (E, φ) to (E 0 , φ0 ) in the C ∗ -ideal D(A, J; B) will
be denoted by Dφ,φ0 (A, J; E, E 0 ; B) (if (E 0 , φ0 ) = (E, φ) then it is also denoted by
Dφ (A, J; E; B)) (cf. [9]) .
p
Theorem 3.2. Let B be a σ-unital C ∗ -algebra and let 0 → I → A → A/I → 0 be
a proper sequence of separable C ∗ -algebras in SG . Then the sequence of groups
. ..
∂
→ Kan (Rep(A/J, B)) → Kan (Rep(A, B)) → Kan (Rep(J, B)) −
→
∂
−
→ Kan−1 (Rep(A, B)) → ... (3.1)
is exact, for all n ∈ Z.
Proof. Consider the short exact sequence of C ∗ -categories and of a C ∗ -ideal
0 → D(A, J; B) → Rep(A, B) → Rep(A, B)/D(A, J; B) → 0.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
206
According to Proposition 2.4, one has a two-sided long exact sequence
... → Kan (D(A, J; B)) → Kan (Rep(A, B)) →
∂
→ Kan (Rep(A, B)/D(A, J; B)) → Kan−1 (D(A, J; B)) → ... (3.2)
of abelian groups. According to Theorems 3.5 and 3.8, replacements
Kan (Rep(A; B)/D(A, J; B)) by Kan (Rep(J; B))
and
Kan (D(A, J; B) by Kan (Rep(A/J; B)),
ensure exactness of the two-sided long exact sequence (3.1). Thus it suffices to prove
Theorems 3.5 and 3.8, which will be done in the next part of this section.
3.1. On the Isomorphism Kan (Rep(A; B)/D(A, J; B)) ≈ Kan (Rep(J; B))
Let (E, φ) be an object in Rep(A, B) and let j : J → A be the natural equivariant
inclusion. There is a ∗-functor induced by the natural inclusion j
j : Rep(A; B) → Rep(J; B)
(3.3)
defined by the assignments (E, φ) 7→ (E, φj) and x 7→ x.
The following trivial lemma is used in the next proposition.
Lemma 3.3. Let A and B be additive C ∗ -categories and let F : A → B be an
additive ∗-functor. Then F is a ∗-isomorphism if and only if F is bijective on
objects and the induced ∗-homomorphisms of C ∗ -algebras Fa : L(a) → L(f (a)) are
∗-isomorphisms for all objects a in A.
The following proposition is a slight generalization of the similar result in [9]).
Proposition 3.4. The canonical ∗-functor 3.3 maps D(A, J; B) to D(J, J; B) and
the induced ∗-functor
ξ : Rep(A; B)/D(A, J; B) → Rep(J; B)/D(J, J; B)
(3.4)
∗
is an isomorphism of C -categories.
Proof. (cf. [9]) By lemma 3.3 it suffices to show that for any object (E, φ) the
∗-homomorphism of C ∗ -algebras
ξJ,φ : Dφ (A; E; B)/Dφ (A, J, E; B) → Dφ·j (J, E; B)/Dφ·j (J, J, E; B)
is a ∗-isomorphism. It is easy to show that ξJ,φ is a monomorphism. To show that
ξJ,φ is an epimorphism, take x ∈ Dφ·j (J, E; B) and let E1 be the G-C ∗ -algebra in
L(E) generated by φ(J) ∪ K(E); let E2 be the separable G-C ∗ -algebra generated
by all elements of the form [x, φ(y)], y ∈ J; and let F be the G-invariant separable
linear space generated by x and φ(A). One has
• E1 · E2 ⊂ K(E), because φ(b)[φ(a), x] ∼ [φ(ba), x] ∈ K(E), a ∈ A, b ∈ J,
• [F, E1 ] ⊂ E1 ,
because [x, φ(J)] ⊂ K(E) and [φ(A), φ(J)] ⊂ φ(J).
From a technical theorem by Kasparov it follows that there exists a positive
G-invariant operator X such that
Journal of Homotopy and Related Structures, vol. 1(1), 2006
207
1. X · φ(J) ⊂ K(E);
2. (1 − X) · [φ(A), x] ⊂ K(E);
3. [x, X] ∈ K(E).
Since [(1 − X)x, φ(a)] = (1 − X)[x, φ(a)] − [X, φ(a)]x, it follows from (2) and
(3) that (1 − X)x ∈ Dφ (A, E; B). In addition, it follows from (2) that Xx ∈
Dφ·j (J, J, E; B), and so that the image of (1−X)x in Dφ·j (J, E; B)/Dφ·j (J, J, E; B)
coincides with the image of x.
Now, we prove the following
Theorem 3.5. Let A be a separable C ∗ -algebra and let B be a σ-unital C ∗ -algebra.
Let J be a closed ideal in A. There exists an essential isomorphism
Kan (Rep(A, B)/D(A, J; B)) ≈ Kan (Rep(J, B))
(3.5)
Proof. According to Proposition 3.4, it suffices to show that the homomorphism
Ka∗ (Rep(J; B)) → Ka∗ (Rep(J; B)/D(J, J; B))
is an isomorphism. The exact sequence
... → Kan (D(J, J; B)) → Kan (Rep(J, B)) →
∂
→ Kan (Rep(J, B)/D(J, J; B)) → Kan−1 (D(J, J; B)) → ... (3.6)
shows that it suffices to show Ka∗ (D(J, J; B)) = 0. According to Kasparov’s stabilization theorem, one concludes that the full subcategory RepHB (J; B) on all objects
of the form (HB , ϕ) is a cofinal subcategory in Rep(J; B), where HB is Kasparov’s
universal Hilbert B-module [23]. Note that the canonical isometry
i1 : HB → HB ⊕ HB
in the first summand is in Dφ,φ⊕0 (J; HB , HB ⊕ HB ; B). So it induces inner homomorphism
ad(i1 ) : Dφ (J, J; HB ; B) → Dφ⊕0 (J, J; HB ⊕ HB ; B).
Consider the sequence of ∗-homomorphisms
Dφ (J, J; HB ; B) → Dφ⊕φ (J, J; HB ⊕ HB ; B) ⊂ Dφ⊕0 (J, J; HB ⊕ HB ; B),
(3.7)
where the inclusion is given by x 7→ x. If the first arrow is induced by the inclusion
ι1 : HB → HB ⊕ HB into the first summand, then the composite is ad(i1 ). If
the first arrow is induced by the inclusion ι2 : HB → HB ⊕ HB into the second
summand, one obtains a homomorphism λ. Since Kna (ad(ι1 )) = Kna (ad(ι2 )), one has
Kna (ad(i1 )) = Kna (λ). On the other hand, the homomorphism λ is the composite of
∗-homomorphisms of C ∗ -algebras
Dφ (J, J; HB ; B) → D0 (J, J; HB ; B) → Dφ⊕0 (J, J; HB ⊕ HB ; B),
defined by the assignments
µ
x 7→ x and x 7→
0
0
0
x
¶
.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
208
Note that one has D0 (J, J; HB ; B) ' M (J ⊗ KG ). It is a well-known fact that
the latter algebra has trivial algebraic K-groups. If we apply K-functors then the
homomorphism corresponding to λ will be zero. Now, as a consequence we have the
following. Let α ∈ Kna (Dφ (J, J; HB ; B)) represent an element in Kan (D(J, J; B)).
Since
α = Kna (ad(i1 )(α)) = Kna (λ(α)) = 0,
one concludes that the class of α in Ka∗ (D(J, J; B)) is zero. Therefore
Ka∗ (D(J, J; B)) = 0.
3.2. On the Isomorphism Kan (Rep(A/J; B)) ' Kan (D(A, J; B))
Let
j
q
0→J −
→A−
→ A/J → 0
be a proper exact sequence and let σ : A/J → A be a completely positive and
contractive (equivariant) section.
Let (E, φ) be an object in Rep(A; B). A ∗-homomorphism
µ
¶
ψ11 ψ12
ψ=
: A/J → L(E ⊕ E 0 ))
ψ21 ψ22
will be called a σ-dilation for φ if ψ11 (a) = φ(σ(a)), where E 0 is a right Hilbert Bmodule. By generalized Stinespring’s theorem there exists a σ-dilation for φ, where
σ : A/J → A is a completely positive and contractive section [23].
Lemma 3.6. Let ψ be a σ-dilation for φ. Then
1. ψ12 (a∗ ) = ψ21 (a)∗ ;
2. for any a, b ∈ A/J there exists a j ∈ J such that ψ12 (a)ψ21 (b) = φ(j);
3. ψ12 (a)x and xψ21 (a) are compact morphisms for any a ∈ A/J and x ∈
Dφ (A, J; B);
4. Let φ : A/J → L(E) be a ∗-homomorphism and let ψ : A/J → L(E ⊕ E 0 ) be
a σ-dilation for φq. There exists a ∗-homomorphism ϕ : A/J → L(E 0 ) such
that
µ
φ
ψ=
0
¶
0
.
ϕ
Proof. The case (1) is trivial, because ψ is a ∗-homomorphism.
The case (2). Since ψ is σ-dilation for φ, one has φ · (σ(ab) − σ(a) · σ(b)) =
ψ12 (a) · ψ21 (b). But j = σ(ab) − σ(a) · s(b) ∈ J. Therefore ψ12 (a) · ψ21 (b) = φ(j).
The case (3). If x ∈ Dφ (A, J; B) then, by definition, xφ(j) and φ(j)x are compact morphisms for any j ∈ J. According to cases (1) and (2), one has xψ12 (a) ·
∗
ψ12
(a)x∗ = xφ(j 0 )x∗ for some j 0 ∈ J. Thus xψ12 (a) · ψ12 (a∗ )x∗ is a compact morphism. Therefore xψ12 (a) and ψ21 (a)x (= (x∗ ψ12 (a∗ ))∗ ) are compact morphisms,
too.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
209
The case (4). Since ψ is a σ-dilation for φq,
µ
¶
φ ψ12
ψ=
.
ψ21 ϕ
According to case (2), for any a, b ∈ A/J there exists j ∈ J such that ψ12 (a)ψ21 (b) =
∗
φq(j) = 0. Applying the case (1), one has ψ12 (a)ψ12
(a) = 0. Therefore ψ12 (a) = 0
and ψ21 (a) = 0 for all a ∈ A/J and ϕ is a ∗-homomorphism.
The following lemma is used in Theorem 3.8.
Lemma 3.7. Let (E, φ) be an object in Rep(A, B) and let ψ : A/J
 →
µ
¶
x11
x11 x12
be a σ-dilation for φ. Then a map x =
7→ x0 =  0
x21 x22
x21
defines a ∗-monomorphism
0
L(E ⊕ E
)
0 x12
0 0 
0 x22
ρ : M2 (Dφ (A, J, E ⊕ E; B)) → Dψ·q⊕φ (A, J, E ⊕ E 0 ⊕ E; B).
(3.8)
0
0
Proof. It suffices to show that x ∈ Dψ·q⊕φ (A, J, E ⊕ E ⊕ E; B). By assumption
one has
(φ(a) ⊕ φ(a))x − x(φ(a) ⊕ φ(a)) ∈ K(E ⊕ E),
for any a ∈ A, and
(φ(b) ⊕ φ(b))x ∈ K(E ⊕ E),
x(φ(b) ⊕ φ(b)) ∈ K(E ⊕ E) for any b ∈ J. It implies that
φ(a)xmn − xmn φ(a) ∈ K(E),
φ(b)xmn ∈ K(E),
0
xmn φ(b) ∈ K(E),
0
a ∈ A and b ∈ J. Then (ψ(q(a)) ⊕ φ(a)) · x − x · (ψ(q(a)) ⊕ φ(a)) =


x11 ψ11 (q(a)) − ψ11 (q(a))x11 x11 ψ12 (p(a)) x12 φ(a) − ψ11 (q(a))x12
.
ψ21 (q(a))x11
0
ψ21 (q(a))x12
=
x21 ψ11 (q(a)) − φ(a)x21
x21 ψ12 (q(a))
x22 φ(a) − φ(a)x22
By Lemma 3.6 (3), the morphisms ψ21 (q(a))x11 , x11 ψ12 (q(a)), x21 ψ12 (q(a)) and
ψ21 (q(a))x12 are compact. Using the fact that φ(a) − ψ11 (q(a)) ∈ φ(J), one has
(ψ(q(a)) ⊕ φ(a)) · x0 − x0 · (ψ(p(a)) ⊕ φ(a)) ∈ K(E ⊕ E 0 ⊕ E), a ∈ A.
To show that (ψ(q(b)) ⊕ φ(b)) · x0 and x0 · (ψ(q(b)) ⊕ φ(b)) are in K(E ⊕ E 0 ⊕ E)
when b ∈ J, note that (ψ(q(b)) ⊕ φ(b)) · x0 and x0 · (ψ(q(b)) ⊕ φ(b)) are equal to




0
0
0
0 0 x12 φ(b)

 and  0 0

0
0
0
0
φ(b)x21 0 φ(b)x22
0 0 x22 φ(b)
respectively. They are compact morphisms because each entry of matrices is a compact morphism.
The category Rep(A, B) may be identified with the full C ∗ -subcategory
D (A, J; B) in D(A, J; B), on the objects all pairs of the form (E, φ · q), considering (E, φ) as an object in Rep(A/J; B). Let
(q)
ε : D(q) (A, J; B) ,→ D(A, J; B)
Journal of Homotopy and Related Structures, vol. 1(1), 2006
210
be the natural inclusion.
j
q
Theorem 3.8. Let 0 → J −
→A−
→ A/J → 0 be a proper exact sequence of separable
C ∗ -algebras. Then the induced homomorphism
Γn : Kan (Rep(A/J; B)) → Kan (D(A, J; B))
is an isomorphism for all n ∈ Z.
Proof. According to the discussion above, it suffices to show that ε induces an
isomorphism
εn : Kan (D(q) (A, J; B)) → Kan (D(A, J; B)),
for all n ∈ Z.
(1). εn is a monomorphism.
Let (E, φ · q) be an object in D(q) (A, J; B) and suppose that the class of an
element
y ∈ Kna (Dφ·q (A, J; E, B))
in Kan (D(A, J; B)) is zero. This means that there exists an isometry
s : (E, φ · q) → (E 0 , ψ)
in Rep(A; B) such that Ad(s)n (y) = 0 in Kna (Dψ (A, J; E 0 , B)).
According to Lemma 3.6, it is easy to show that
µ
¶
s 0
s0 =
: (E ⊕ HB , η · q) → (E 0 ⊕ HB , η 0 · q)
0 0
is an isometry in Rep(A/J, B), where η and η 0 are σ-dilations of φ · q and ψ respectively. By Lemma 3.6 η has form
µ
¶
φ 0
η=
,
0 χ
where χ is a ∗-homomorphism from A to L(HB ). There is a homomorphism
µ
defined by x 7→
ν : Dψ (A, J; E 0 , B) → Dη0 q (A, J; E 0 ⊕ HB , B),
¶
x 0
, and an isometry
0 0
µ
¶
φ 0
)
i1 : (E, φ) → (E ⊕ HB , η =
0 χ
(inclusion into the first summand). It is clear that Ad(s0 i1 ) = νAd(s). Therefore
Ad(s0 i1 )n (y) = 0 in Dη0 q (A, J; E 0 ⊕ HB , B) and in Kan (D(q) (A, J; B)) too. This
means that the class of y in Kan (D(q) (A, J; B)) is zero.
(2). εn is an epimorphism.
Let an element in Kan (D(A, J; B)) be represented by an element
x ∈ Kna (Dφ (A, J; E, B)). Consider the ∗-homomorphism
θ : Dφ (A, J; E; B) → Dψq (A, J; E ⊕ HB , B)
Journal of Homotopy and Related Structures, vol. 1(1), 2006
given by
µ
x 7→
x 0
0 0
211
¶
,
where ψ is a σ-dilation of φ.
Let us show that classes of elements θn (x) and x in Kan (D(A, J; B)) coincide.
Let ϑ be the composite
Ad(iE⊕H )
θ
B
Dφ (A, J; E; B) −
→ Dψq (A, J; E ⊕ HB ; B) −−−−−−−
→ Dψq⊕φ (A, J; E ⊕ HB ⊕ E; B),
where the second arrow is induced by the isometry into the first two summands.
It is clear that the ∗-homomorphism ϑ coincides with the ∗-homomorphism defined
by


x 0 0
x 7→  0 0 0  .
0 0 0
On the other hand this homomorphism may be interpreted as the composite
i
ρ
1
Dφ (A, J; E, B) −→
M2 (Dφ (A, J; E, B)) −
→ Dψ·q⊕φ (A, J; E ⊕ HB ⊕ E, B)
µ
¶
x 0
where i1 is given by x 7→
and ρ is defined by
0 0


µ
¶
x11 0 x12
x11 x12
x=
7→ x0 =  0 0 0  ,
x21 x22
x21 0 x22
as in Lemma 3.7. Consider another homomorphism η—the composite of the sequence of ∗-homomorphisms
i
ρ
2
Dφ (A, J; E, B) −→
M2 (Dφ (A, J; E, B)) −
→ Dψ·q⊕φ (A, J; E ⊕ HB ⊕ E, B)
µ
¶
0 0
where i2 is given by x 7→
. Since (i1 )n = (i2 )n , n ∈ Z, one has (ϑ)n =
0 x
(η)n . But (η)n = (ad(i3 ))n , n ∈ Z, where i3 : E → E ⊕ HB ⊕ E is an isometry
in the third summand. Therefore, classes of the elements ϑn (x), (ad(i3 ))n (x), x
and θn (x) in Kan (D(A, J; B)) coincide. Therefore the class of the element θn (x) in
Kan (D(q) (A, J; B)) is the desired element.
4.
Stability Property of Kan (Rep(−; B))
Everywhere below in this section, K is the C ∗ -algebra of compact operators on a
countable generated Hilbert space H considered as an object of SG via trivial action
of the compact group G.
Let p ∈ K be a rank one projection and let A be a C ∗ -algebra in SG ; let eA :
A → A ⊗ K be the ∗-homomorphism defined by a 7→ a ⊗ p, a ∈ A. Then one has
the induced functor
e∗A : Rep(A ⊗ K; B) → Rep(A; B),
(4.1)
Journal of Homotopy and Related Structures, vol. 1(1), 2006
212
defined by assignments (E, ϕ) 7→ (E, ϕeA ) (on objects) and x 7→ x (on morphisms).
There is a ∗-functor
εA : Rep(A; B) → Rep(A ⊗ K; B)
defined by assignments (E, φ) 7→ (E ⊗k H, φ ⊗ idK ) (on objects) and f 7→ f ⊗ idH
(on morphisms). Indeed, let (E, φ) be an object in Rep(A; B). One has the induced
∗-homomorphism
φ ⊗ idK : A ⊗ K → L(E ⊗k H).
Let f : (E, φ) → (E 0 , φ0 ) be a morphism in Rep(A; B), i. e. f φ(a) − φ0 (a)f ∈
K(E, E 0 ), a ∈ A. Then
(f ⊗ idH )((φ ⊗ idK )(a ⊗ κ)) − ((φ ⊗ idK )(a ⊗ κ))(f ⊗ idH ) =
= (f φ(a) − φ(a)f ) ⊗ κ ∈ K(E ⊗k H, E 0 ⊗k H)
(4.2)
for all a ∈ A, κ ∈ K.
Now, in view of Remark 3.1 stability property of Kan (Rep(−; B)) may be formulated as follows.
Theorem 4.1. For any rank one projection p ∈ K and any C ∗ -algebra A in SG the
homomorphism
a ∗
a
a
eA
n = Kn (eA ) : Kn (Rep(A ⊗ K; B)) → Kn (Rep(A; B)),
induced by the functor 4.1 is an isomorphism.
a
a
Proof. Let εA
n : Kn (Rep(A; B)) → Kn (Rep(A ⊗ K; B)) be the homomorphism
A
induced by the functor ε . It is easy to verify that the family {εA
n } is a natural
transformation from the functor Kan (Rep(−; B)) to Kan (Rep(− ⊗ K; B)). Therefore,
the following diagram
Kan (Rep(A ⊗ K; B))


εA⊗K
y
n
eA
−−−n−→
Kan (Rep(A; B))

A
yεn
Kan (Rep(A ⊗ K ⊗ K; B)) −−−−→ Kan (Rep(A ⊗ K; B))
eA⊗K
n
commutes, and it shows that for our purposes it suffices to verify that
A
eA
,
n εn = idKa
n (Rep(A;B))
(4.3)
for all A ∈ SG . Indeed, first note that
1. The equality 4.3 shows that eA
n is an epimorphism.
2. Since Kan (Rep(− ⊗ K; B)) is a stable functor, according to the identity 4.3, one
easily shows that εA⊗K
and eA⊗K
are isomorphisms (cf. Proposition 10.6 in [33]).
n
n
A
Therefore en is a monomorphism.
Checking the equality 4.3. We construct a useful isometry σE : E → E ⊗k H,
for any countably generated B-module E, which is a morphism from (E, φ) to
(E ⊗k H, (φ ⊗ idK )eA ).
Journal of Homotopy and Related Structures, vol. 1(1), 2006
213
Choose y ∈ H so that p(y) = y and ||y|| = 1 and consider a B-homomorphism σE
given by x 7→ x ⊗ y. For any z ∈ H there exists λz ∈ k determined uniquely by the
∗
equation p(z) = λz y. Define σE
by the map x ⊗ z 7→ λz x. The B-homomorphism
∗
∗
∗
σE is adjoint to σE . Since σE σE (x) = σE
(x ⊗ y) = x, one concludes that σE is an
isometry. Since σE φ(a) = ((φ ⊗ idK )eA (a))σE , the isometry σE is a morphism from
(E, φ) into (E ⊗k H, (φ ⊗ idK )eA ).
Consider restriction of e∗A ε to Dφ (A; E; B). We have a ∗-homomorphism
(e∗A ε)E : Dφ (A; E; B) → D(φ⊗id)eA (A; E ⊗k H; B)
(4.4)
mapping x ∈ Dφ (A; E; B) to x ⊗ idH ∈ D(φ⊗id)eA (A; E ⊗k H; B). But
(σE x)(z) = (σE )(x(z)) = x(z)⊗y = ((x⊗p)σE )(z) for any x ∈ Dφ (A; E; B), z ∈ E.
∗
∗
Since σE σE
= idE ⊗ p, one concludes that σE xσE
= x ⊗ p. Therefore
∗
(e∗A ε)E (x) = σE xσE
+ x ⊗ (1 − p).
(4.5)
Let ψ(x) = (e∗A ε)E (x), ψ0 (x) = x ⊗ p and ψ1 (x) = x ⊗ (1 − p). Then ψ0 and ψ1 are
∗-homomorphisms, ψ = ψ0 + ψ1 and
ψ0 (x)ψ1 (x) = ψ1 (x)ψ0 (x) = 0.
(4.6)
Here we show that e∗A ε induces the identity homomorphism of the group
Kan (Rep(A, B)) onto itself. Indeed, choose an element r ∈ Kan (Rep(A, B)).
By definition of Kan -groups the element r is represented by an element
rφ ∈ Kna (Dφ (A; E; B)). Then the element Kan (e∗A ε)(r) is represented by the element
Kna ((e∗A ε)E )(rφ ) = Kna (ψ0 + ψ1 )(rφ ).
(4.7)
Since Kna is an additive functor, according to 4.6 and Lemma 2.1.18 in ([11]), it
follows that
Kna (ψ0 + ψ1 )(rφ ) = Kna (ψ0 )(rφ ) + Kna (ψ1 )(rφ ).
Since ψ0 = ad(sE ), the class of Kna (ψ0 )(rφ ) is equal to the class of rφ . Thus the
proof will be completed if we show that the class of Kna (ψ1 )(rφ ) in Kan (Rep(A; B))
is zero. Indeed, let
s : E ⊗ H → (E ⊗ H) ⊕ HB )
be the isometry defined by e ⊗ h 7→ e ⊗ h ⊕ 0. Then the ∗-homomorphism ad(s)ψ1
may be factored through the ∗-homomorphism
Dφ (A; E; B) → D0 (A; E ⊗ (1 − p)H ⊕ HB ; B).
But, according to Kasparov’s stabilization theorem, one has
D0 (A; E ⊗ (1 − p)H ⊕ HB ; B) ≈ L(HB ).
Since Kna (L(HB )) = 0, one concludes that classes of Kna (ψ1 )(z) and
Kna (ad(s)(ψ1 ))(z) are equal to zero in Kan (Rep(A; B)). Thus the homomorphism
Kan (eA ε) is the identity.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
5.
214
On the Isomorphism Ka0 (Rep(−; B)) ' KK−1 (−; B)
First we recall the definition of Kasparov’s group K 1 (A, B), which will be denoted
by KK−1 (A, B), where A and B are trivially graded C ∗ -algebras with actions of a
second countable compact group.
Consider a triple (ϕ, E; p), where E is a trivially graded countably generated right
B-module, ϕ : A → LB (E) is a ∗-homomorphism and p ∈ LB (E) is an invariant
element so that
pϕ(a) − ϕ(a)p ∈ KB (E),
(p∗ − p)ϕ(a) ∈ KB (E), (p2 − p)ϕ(a) ∈ KB (E)
(5.1)
for all a ∈ A. Such a triple will be called a Kasparov-Fredholm A, B- module. If all
left parts in 5.1 are zero, then such a triple is said to be degenerate.
Define the sum of Kasparov-Fredholm A, B-modules by the formula
(ϕ, E; p) ⊕ (ϕ0 , E 0 ; p0 ) = (ϕ ⊕ ϕ0 , E ⊕ E 0 ; p ⊕ p0 ).
Consider the equivalence relations:
• (Unitary isomorphism) A, B-modules (ϕ, E; p) and (ϕ0 , E 0 ; p0 ) will be said to
be unitarily isomorphic if there exists a unitary isomorphism u : E → E 0 such
that
uϕ(a)u∗ = ϕ0 (a), upu∗ = p0
for all a ∈ A.
• (Homology) A, B-modules (ϕ, E; p) and (ϕ0 , E; p0 ) will be said to be homologous if
p0 ϕ0 (a) − pϕ(a) ∈ KB (E)
for all a ∈ A.
Simple checking shows that the equivalence relations defined above are well behaved
with respect to sum.
Let E 1 (A, B) be the abelian monoid of classes of A, B-modules with respect to the
equivalence relation generated by the unitary isomorphism and homology. Denote
by D1 (A, B) the submonoid of E 1 (A, B) consisting of only those classes which are
classes of all degenerate triples. By definition
E 1 (A, B) = E 1 (A, B)/D1 (A, B).
Using the Kasparov stabilization theorem, one easily shows that the definition of
E 1 (A, B) coincides with Kasparov’s original definition of E 1 (A, B) which is isomorphic to KK−1 (A, B) by lemma 2 of section 7 of [24].
Recall that objects of Rep(A, B), by definition, have form (ϕ, E; p), where p :
(ϕ, E) → (ϕ, E) is a projection in the category Rep(A; B). More precisely,
ϕ(a)p − pϕ(a) ∈ KB (E), p∗ = p, p2 = p.
A unitary isomorphism s : (ϕ, E; p) → (ψ, E 0 , q) in Rep(A, B) is a usual partial
isometry s : E → E 0 such that
sϕ(a) − ψ(a)s ∈ KB (E, E 0 ), s∗ s = p, ss∗ = q.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
215
Let Ẽ 1 (A, B) be the abelian monoid of unitary isomorphism classes of objects in
Rep(A; B). According to Lemma 1.1, one easily checks that the Grothendieck group
of Ẽ 1 (A, B) may be identified with K0 (Rep(A, B)) (cf. [16]).
There is a natural homomorphism
λ1 : K0 (Rep(A, B)) → E 1 (A, B),
defined by the map (ϕ, E; p) 7→ (ϕ, E; p). Indeed, let s : (ϕ, E; p) → (ϕ0 , E 0 ; p0 ) be a
unitary isomorphism in Rep(A; B). Consider the isomorphism in Rep(A; B)
s̄ : (ϕ ⊕ ψ, E ⊕ E 0 ) → (ψ ⊕ ϕ, E 0 ⊕ E),
where
µ
s̄ =
s
1 − s∗ s
1 − ss∗
s
¶
.
It is clear that (ϕ ⊕ ψ, E ⊕ E 0 , p̄) is isomorphic to (s(ϕ ⊕ ψ)s∗ , E 0 ⊕ E, q̄), which is
homologous to (ψ ⊕ ϕ, E 0 ⊕ E, q̄), with
µ
¶
µ
¶
p 0
1 0
0
p̄ =
and p̄ =
.
0 1
0 p0
This means that classes of (ϕ, E; p) and (ϕ0 , E 0 ; p0 ) coincide in E 1 (A, B).
Let RepJ (A, B) = Rep(A, B)/D(A, J; B) be the idempotent-complete
C ∗ -category universally obtained from the category Rep(A, B)/D(A, J; B)) (see
section 1). Let (ϕ; E; p) be a Kasparov-Fredholm A, B-module. Then p defines a
projector ṗ in the category RepA (A, B). Thus the triple (ϕ; E; ṗ) is an object in
RepA (A, B).
There is a well-defined homomorphism
µ : E 1 (A, B) → K0 (RepA (A, B))
defined by (ϕ; E; p) 7→ (ϕ; E; ṗ). This is checked below.
We recall definition of operatorial homotopy:
• (Operatorial homotopy) An A, B-module (ϕ, E; p) is operatorially homotopic
to (ϕ, E; p0 ) if there exists a continuous map pt : [0; 1] → LB (E) such that
(ϕ, E; pt ) is an A, B-module for any t ∈ [0; 1].
If (ϕ, E; p) is homologous to (ψ, E; q), then (ϕ, E; p) ⊕ (ψ, E; 0) is operatorially
homotopic to (ϕ, E; 0) ⊕ (ψ, E; q). Indeed, the desired homotopy is defined by the
formula
µµ
¶
µ
¶¶
1
ϕ 0
p tpq
, E ⊕ E,
, t ∈ [0; ∞]
0 ψ
tqp t2 q
1 + t2
˙ 0 and 0 ⊕
˙ q are homotopic. Then,
(cf. section 7 in [24]). Thus the projections p ⊕
using Lemma 4 from section 6 in [24], one concludes that the objects (ϕ, E; ṗ) ⊕
(ψ, E; 0̇) and (ϕ, E; 0̇) ⊕ (ψ, E; q̇) are unitarily isomorphic objects in RepA (A, B).
Let (ϕ, E; p) be unitarily isomorphic to (ψ, E; q). Then (ϕ, E; ṗ) is isomorphic to
(ψ, E; q̇) in the category RepA (A, B). Therefore µ is well-defined.
We are now ready to prove the following theorem.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
216
Theorem 5.1. The natural homomorphism
λ1 : K0 (Rep(A, B)) → E 1 (A, B) ' KK−1 (A, B)
is an isomorphism.
Proof. The homomorphism λ1 is an epimorphism. Indeed, let (ϕ, E; p) be a Kasparov-Fredholm A, B-module. Applying techniques of the Lemmata 17.4.2-17.4.3 in
[2], one can suppose that p∗ = p and ||p|| 6 1. Then it is equivalent to (ϕ ⊕ 0, E ⊕
E; p0 ), where
p
µ
¶
p
p − p2
0
p
p =
.
p − p2
1−p
Simple checking shows that p0 is a projection and (ϕ ⊕ 0, E ⊕ E; p0 ) is an object in
Rep(A, B). To show that λ1 is a monomorphism, consider the commutative diagram
λ
K0 (Rep(A, B)) −−−1−→


ky
E 1 (A, B)

µ
y
ξ
K0 (Rep(A, B)) −−−−→ K0 (RepA (A, B)).
By Theorem 3.5, ξ is an isomorphism. Therefore λ1 is a monomorphism.
Acknowledgements
The work has synthesized ideas of many mathematicians. The author would like
to thank, especially those of, Cuntz, Higson, Karoubi, Kasparov, Pedersen, Quillen,
Rosenberg, Suslin, Weibel, Wodzicki.
It is great pleasure to acknowledge to J. Rosenberg and the referee for the valuable
comments.
The author was partially supported by INTAS grant 00 566, INTAS grant 0351-3251 and GRDF grant GEM1-3330-TB-03.
References
[1] Bass H. Algebraic K-theory, Benjamin, New York (1968).
[2] Blackadar B. K-theory for Operator Algebras, M.S.R.I. Publ. 5, SpringerVerlag, (1986).
[3] P. Chez, R. Lima and J. Roberts, W ∗ -categories, Pacific J. Math. 120
(1985), No 1, 79-109.
[4] Connes, A. and Higson N. Deformations, morphisms asymptotiques et Ktheorie bivariante, C.R. Acad. Sci. Paris, Sér. I Math. 313 (1990), 101106.
[5] Cuntz J. K-theory and C ∗ -algebras, Lecture Notes In Math. 1046, 55-79.
[6] Cuntz J. A New Look at KK-theory K-theory, v.1 (1987), 31-51.
[7] Cuntz J. Skandalis G. mapping cones and exact sequences in KK-theory
J. operator theory, v.15, (1986), 163-180.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
217
[8] Higson, N.: Categories of fractions and excision in KK-theory, J. Pure
Appl. Algebra 65 (1990), 119-138.
[9] Higson N. C ∗ -algebra extension theory and duality , J.Funct. Anal., v.129,
(1995), 349-363.
[10] Higson N. A characterization of KK -theory, Pacific J. Math., v.126 No.
2, (1987), 253-276.
[11] Higson N. Algebraic K-theory of stable C ∗ -algebras, Adv. Math., v.67,
(1988) 1-140.
[12] Higson N.,Pedersen E.K., Roe J. C ∗ -algebras and controlled topology Ktheory 11, (1997) 209-239.
[13] Houghton-Larsen T.G. Thomsen K. Universal (Co)Homology Theories, KTheory 16, (1999), 1–27.
[14] Inassaridze H. Algebraic K-theory, Kluwer Acad. Publ., Dordrecht Boston
London, v.311, (1995).
[15] Inassaridze H. Kandelaki T. Smooth K-theory of locally convex algebras
(to appear).
[16] Kandelaki T. KK-theory as the K-theory of C ∗ -categories, Homotopy,
Homology and Applications, v.2, No. 10, (2000), 127-145.
[17] Kandelaki T. Algebraic and Topological bivariant KK-theories of C ∗ algebras and their isomorphism. Proc. A.Razmadze Math. Inst., v. 126,
(2001), 110-113.
[18] Kandelaki T. Multiplier and Hilbert C ∗ -categories Proc. A.Razmadze
Math. Inst., v. 127, (2001), 89-111.
[19] Kandelaki T. Karoubi-Villamayor K-theory, weakly stable C ∗ -categoroids,
and KK-theory Georgian Math. J. v. 11, n. 2, (2004), 283-299.
[20] Kandelaki T. Algebraic
arXiv.org.(April, 2003)).
K-theory
view
on
KK-theory(www.
[21] Karoubi M., Algebres de Clifford et K -théorie. Ann. Sci. Ècole norm. Sup.,
t.1., f.2(1968)161-268.
[22] Karoubi M. K-Theory, Compr. Studies Math. 226., Springer-Virlag,
(1978).
[23] Kasparov G. Hilbert C ∗ -modules: Theorems of stinespring and voiculescu,
J. Operator theory, v. 4, (1980), 133-150.
[24] Kasparov G., The operator K-functor and extension of C ∗ -algebras,
Izv.A.N. SSSR, v.44, (1980), 571-636. (in Russian).
[25] Meyer R. Nest R. The Baum-Connes conjecture via localisation of categories, Topology, vol. 45 (2006), pp. 209-259.
[26] Mitchener P. KK-Theory of C*-Categories and the Analytic Assembly
Map, K-Theory 26, (2002), 307-344.
[27] Palmer T.W. Real C ∗ -algebras, Pacific J. Math,.v.35 (1970),195-204.
Journal of Homotopy and Related Structures, vol. 1(1), 2006
218
[28] Pedersen E.K. Weibel C.A. A nonconnective delooping of algebraic Ktheory Lect. Notes in Math. 1126, Spinger Verlage, Berlin-New York,
(1984) 540-557.
[29] Pedersen E.K. Weibel C.A. K-theory homology of spaces Lecter Notes in
Math. 1370,, Springer Verlage, Berlin-New York, (1989) 341-361.
[30] Quillen D. Higher Algebraic K-theory, 1, Lect. Notes in Math., v.341
(1973), 77-139.
[31] Rosenberg J. K and KK: Topology and Operator Algebras Proccedings of
Symposia in Pure Mathematics, v.51, p.1, (1990), 445-479.
[32] Skandalis G. Some remarks on Kasparov theory, J. Funct. Anal. v. 56,
(1984), 337-347.
[33] Suslin A., Wodzicki M. Excision in algebraic K-theory, Ann. Math. v.136,
No.1, (1992), 51-122.
This article may be accessed via WWW at http://jhrs.rmi.acnet.ge
Tamaz Kandelaki
nichbisi@yahoo.com
A. Razmadze Mathematical Institute,
1, M. Aleksidze St., 380093, Tbilisi,
Georgia